<p>On-demand instant delivery has increasingly become an integral component of urban logistics systems. Yet, the dynamic mobility pattern of the vast rider fleet introduces significant sustainability challenges. Mitigating these impacts hinges on adaptive transport management grounded in the delivery system’s inherent daily regularities; however, this dynamic mobility landscape remains largely underexplored. Here, using a large-scale dataset from Beijing, we construct time-evolving instant delivery networks and employ a cross-time-layer community detection method to track the hourly evolution of their dynamic community structures. Our analysis identifies 160 distinct communities and profiles their spatiotemporal life cycles, from emergence to dissolution. By examining node variability within these dynamic communities and linking this variability to spatial factors through a machine learning model, we find that the number of pickups and the presence of shopping malls significantly enhance community stability, while building area, working population, and functional diversity increase variability, highlighting mobility patterns shaped by the tension between stable supply and dynamic demand. These findings reveal the underlying structure of delivery mobility, providing a data-driven framework for a crucial policy shift from static rules to adaptive management. We demonstrate this transition by strategies such as allocating delivery space based on identified temporal rhythms and optimizing fleet operations within the boundaries of these dynamic communities, ultimately fostering more efficient and sustainable urban logistics solutions.</p>